LAUSR

Abstract. The presence of haze degrades the quality of a captured image. The aerosols in the atmosphere cause scattering of the incident light, and this phenomenon is observed on the captured image as well, where some regions appear grayish and colors in those regions appear faded. To improve the quality of the hazy images, we propose a joint fusion and restoration model that sufficiently enhances the contrast of the hazy image while preserving its mean brightness. The fusion model utilizes images of various exposures generated by a modified Gamma correction model. The images for fusion are selected using some selection criteria and fused in a multi-resolution decomposition scheme. Three haze-sensitive weight maps corresponding to some statistical property of haze namely, saliency, illumination, and luminance gradient are constructed. The hazy image formation model is then used and dehazing is performed based on the dark channel prior assumption. The proposed algorithm does not consider the estimation of airlight vector which seldom cause over-saturation defect, instead a mean brightness preservation model has been applied. The variety of experiments demonstrate the significance of the proposed method.